Every talk about site speed reaches the same slide. Amazon found 100 milliseconds costs 1 percent of revenue. Walmart tied a second of load time to 2 percent of conversions. Somebody writes it down, and the citation never appears.
We went looking for the receipts. All five of the most-quoted numbers are real, in the sense that a named company published them. None went through peer review. Most are old enough to drive.
- 100ms = 1%revenue at Amazon. Greg Linden, conference talk, 2006
- 1s = 2%conversions at Walmart. Velocity conference talk, 2012
- 53%mobile visits abandoned past 3 seconds. Google report, 2016
- +32%bounce probability, 1s to 3s. Google and SOASTA model, 2017
- 0.1s = +8.4%retail conversions. Deloitte and Google, 37 brands, 2020
The Amazon figure traces to Greg Linden, an engineer describing internal A/B tests in a 2006 presentation. A 20-year-old observation about a website that no longer exists in any meaningful sense. The Walmart number comes from a conference talk. The Google numbers are vendor reports from a company that sells ads, published to persuade sites to get faster.
The Deloitte study from 2020 is the sturdiest of the set. It isolated a 0.1 second improvement across 37 brands and measured an 8.4 percent lift in retail conversions, with travel higher still. Still vendor research. Google funded it, and Google profits when the web speeds up.
What survives scrutiny is the direction. Five sources, five methodologies, four industries, one conclusion. Faster sites convert better everywhere anyone has looked, and the effect registers at tenths of a second. The magnitude for your site is unknowable from any of these studies, because none of them measured your site.
So quote the numbers with their dates attached. And when a speed project needs a business case, the cheapest honest version is a week of your own field data plotted against your own conversions. Your regression beats Amazon's from 2006.